Data-processing workflows may achieve increased throughput when a large amount of data is processed in a batch, for example, due to a reduced overhead relative to each datum in the batch. However, a large batch size may result in a large latency for processing the batch, for example, due to waiting for the whole batch of data to be available.